EEG dynamic noise floor measurement with stochastic flash A/D converter

https://doi.org/10.1016/j.bspc.2017.07.006Get rights and content

Highlights

  • Noise floor measurement in an EEG system enables high accuracy in wearable devices.

  • Stochastic measurement method enables fast processing and high accuracy.

  • Low frequency noise components are measured over long time periods.

  • Hardware prototype with integrated EEG is tested and measured.

  • 0.004% measurement error enables EEG implementation in every-day surroundings.

Abstract

The noise floor is the measure of the signal at the output of a measurement system, produced by internal and external noise sources. The dynamic noise floor quantifies the effects of non-uniform noise. The main objective is to quantify accurately the noise floor in an EEG system, as the measurement error of the biopotential signals is in the microvolts voltage range. Signals measured by an EEG range down to 0.01 Hz, so measurements require long time and produce large quantities of data that needs to be measured with high accuracy. This paper presents a novel idea for the noise floor quantification using stochastic method of measurement over a long time interval. The accuracy of the method is independent of the input noise type, and it depends only on duration of the measurement interval and the flash A/D converter accuracy. The method is based around the 4-bit Stochastic Flash ADC with fast processing time of recorded data and high precision. A mathematical model of the stochastic measurement results is given. When the length of the measurement interval is 100 s, the relative measurement error falls below 0.004%. Long time of measurement and high precision allow this method to be integrated into the mixed-mode system on a chip, as a part of self-calibration process in a wearable wireless medical monitoring device, such as an EEG. The prototype with the integrated EEG chip is described and its noise floor is measured using the 4-bit Stochastic Flash ADC. In conclusion, the measurement results are analyzed and compared to the product datasheet figures, showing the significance of the presented measurement method which does not depend on the type of measured noise.

Introduction

Electroencephalographic (EEG) measurement is commonly used in medical and research areas [1]. EEG signals are nonstationary, low-level electrical biopotentials (generally less than 300 μV) originating from physiological processes in the brain [1], [2]. The frequencies of these voltages can range from 0.01 to 100 Hz, and their characteristics are highly dependent on the level of activity of the cerebral cortex. From a hardware complexity perspective [2], electroencephalograms have traditionally been the most difficult electrophysiological measurements to acquire.

Standard EEG measurement system consists of electrodes and cables, a conditioning module, a digitizing module and a module for performing data processing, recording and presenting. The microvolt level EEG voltage is subjected to noise, often many times greater than the signal itself. To achieve satisfactory amplification of such low level EEG signal, the conditioning module incorporates amplifying circuits with a high gain (5000–20,000 times), but also implements Driven Right Leg (DRL) method [3] and high-order analog filters with a sharp roll-off to ensure that only the desired signal is detected [4]. EEG data processing has a highly important role, because of the significance of many spectral and nonlinear measures [1], [2].

The low frequency components of the inherent noise impose limitations on the EEG frequency band. Low frequencies of the order of 0.1 Hz are significant when measuring phenomena with very slow shifts of electric potential, such as the contingent negative variation (CNV) and the Bereitschaftspotential (BP). Frequency components of 0.5 Hz are present during slow-wave sleep, corresponding to the ultraslow oscillations measured intracellularly from cortical neurons through layers II to VI, consisting of prolonged depolarizing and hyperpolarizing components. Signal amplitudes of 1–300 μV can be obtained in an EEG, so any noise present in the system can affect the accuracy of the measurement thus producing false results or even masking the entire signal. Low-pass filters can reduce high frequency components of the noise, but frequencies below 100 Hz cannot be filtered as most of the information recorded by an EEG lies in that band. Some specific frequencies, like 50 Hz of the mains power line, are filtered with notch filters for frequencies in a narrow band [5].

If noise is reduced to the acceptable level of at least 10 times lower than the EEG signal, then uncontaminated EEG records can be obtained [1].

Section snippets

Noise floor

When low-level signals are processed in modern communication systems, they often tend to be masked by the noise added by the system itself. Sensitivity, bit-error ratio and noise figure are the system parameters that characterize low-level signal processing ability. The noise figure (NF) of a system is defined as a ratio of its signal-to-noise power ratio at the input to its signal-to-noise power ratio at the output. NF characterizes the entire system and its components, and differentiates one

Denoising

There are several standard mathematical methods of noise reduction (denoising) which are used for the removal of noise present in the EEG signals [18]: a) Principal component analysis (PCA) is a mathematical procedure that transforms a number of correlated variables into a smaller number of uncorrelated variables (principal components), b) Independent Component Analysis (ICA) based denoising where components of many signals are often very sparse so noises in the ICA domain can be removed, c)

EEG noise floor measurement

There are two main methods used for measuring the low-level noise present in any communication system: the direct noise measurement method and the signal generator twice-power method [6]. In the former, all inputs are terminated (grounded) and the output is measured whereas in the latter a two-step process is implemented. The first step is similar to the direct noise measurement method, while in the second step, a well-defined reference signal is brought to the input and the output is measured,

Stochastic flash ADC

The classic data sampling approach takes one very short instant of time Δt and measures the amplitude of the input signal in that point. Finite duration of that time instant determines the sampling frequency fs and the maximum frequency of the input signal fy:fs=2fy=1/Δt

The frequency of sampling must be at least twice the signal frequency, presenting technology problems when high-speed signals are measured. A large number of samples must be taken for the accurate reproduction of low-frequency

Measurement setup

The measurement setup for the dynamic noise floor measurement of an EEG using the 4-bit SFADC is given in Fig. 2. All inputs of the EEG Analog Front-End are grounded and data recording is set to 50 min (30 full cycles of 0.01 Hz signal). Data recording is done in the digital section of an EEG or by a PC connected to the EEG digital communication interface. If the EEG has an analog output, it can be directly connected to the 4-bit SFADC.

If the EEG input is grounded, ideally, the zero signal should

Measurement results

A set of 1000 measurements at the sampling frequency of 250 Hz was performed on the ADS1299 prototype, measuring the peak-to-peak amplitude of the output noise signal, with all inputs grounded. The results of the dynamic noise floor measurements are given in Fig. 3.

A similar graph is given in the datasheet for 10 s period, in 2500 data points with the sampling frequency of 250 Hz.

The graph of the dynamic noise floor given in the datasheet must be scaled to four seconds for correct comparison with

Discussion

The results of the measurements presented in this paper, show that noise in the system is, in most cases, close to the ideal Gaussian white noise. The accuracy of the results depends on the measurement period when SFADC method is used, but is not affected by the noise type present. This method can record any type of noise in the system, unlike some standard measurement methods (e.g., sigma-delta ADC).

Due to the fast Flash ADC structure, the 4-bit SFADC is only bound by the clock and the dither

Conclusion

The 4-bit SFADC, based on the method of measurement over long time interval, can be used to measure the dynamic noise floor of an EEG system, a critical parameter used for the system calibration. Its main benefits are the higher precision measurements and the increased speed of processing large amount of data needed for the accurate quantification of the dynamic noise floor in the (0.01–70) Hz range. The measurement error is less than 0.005%. These features enable fast self-calibration of an EEG

References (38)

  • M. Urekar et al.

    Accuracy improvement of the stochastic digital electrical energy meter

    Measurement

    (2017)
  • J.D. Bronzino

    Principles of electroencephalography

  • J.G. Webster

    Medical Instrumentation Application and Design

    (1998)
  • B.B. Winter et al.

    Driven-right-leg circuit design

    IEEE Trans. Biomed. Eng.

    (1983)
  • M. Kutz

    Standard Handbook of Biomedical Engineering and Design

    (2003)
  • F. Lopes da Silva

    EEG: origin and measurement

  • “Fundamentals of RF and Microwave Noise Figure Measurements”, Application Note 57-1

    (2010)
  • P. Enge et al.

    “Measurements of Man-Made Spectrum Noise Floor”, CR-2004-213551

    (2004)
  • F. Goh, J. Kelly, Effects of Noise Floor, Linearity & Mismatch Error on RF Measurements, Verigy...
  • D. Slepicka

    Noise floor in ADC testing

  • IEEE Std. 1241–2000 − IEEE Standard for Terminology and Test Methods for Analog-to-Digital Converters

    (2016)
  • R.M. Page-Jones

    Notes on the RSGB Observations of the HF Ambient Noise Floor

    (2003)
  • IEEE Std. 1057–2007 − IEEE Standard for Digitizing Waveform Recorders

    (2014)
  • A.S. Zandi et al.

    Scalp EEG acquisition in a low-noise environment: a quantitative assessment

    IEEE Trans. Biomed. Eng.

    (2011)
  • P. Sovilj et al.

    Digital stochastic measurement of a nonstationary signal with an example of EEG signal measurement

    IEEE Trans. Instrum. Meas.

    (2011)
  • Symtop UEA-BZ,...
  • g.tec USBamp,...
  • NeuroScan SanAmps2,...
  • P. Khatwani et al.

    A survey on different noise removal techniques of EEG signals

    Int. J. Adv. Res. Comput. Commun. Eng.

    (2013)
  • Cited by (8)

    • Lesion localization algorithm of high-frequency epileptic signal based on Teager energy operator

      2019, Biomedical Signal Processing and Control
      Citation Excerpt :

      Epilepsy is already a deadly disease which is only second to cerebrovascular disease in neurological diseases [4]. It has great significance to study various abnormal EEG for the diagnosis and treatment of epilepsy [5]. Currently, the research for epilepsy EEG is mostly concentrated in the frequency band below 100 Hz.

    • Stochastic approach for controllable measurement uncertainty in Industry 4.0 applications

      2020, 2020 IEEE International Workshop on Metrology for Industry 4.0 and IoT, MetroInd 4.0 and IoT 2020 - Proceedings
    • Low power stochastic sensor in IoT, IIoT and industry 4.0 environments

      2020, 24th IMEKO TC4 International Symposium and 22nd International Workshop on ADC and DAC Modelling and Testing
    View all citing articles on Scopus
    View full text